Towards activity recognition in smart homes using multimodal data

Recognizing the activities of residents of a Smart Home is an essential prerequisite to derive their intentions and typical behaviors and, subsequently, to adapt the functions of the Smart Home to these. In this paper, we describe an experimental setup for Human Activity Recognition using a special multisensor platform. In the course of the experiment, sensor data and the associated activity labels are acquired and aggregated into a comprehensive dataset. This data will be used to train and validate different machine learning models that analyze the data in real time. These models will be used to perform realtime streaming predictions of activities on an edge device that is located in the home of the user. The goal is to provide activity and context recognition as a basic service for subsequent action routines. Possible use cases can be found in the areas of comfort management and assisted living for, e.g., elderly people.